|
169 | 169 | - MCP benefit score 0.73 for workflow discovery: high context complexity (large repo), moderate cross-file deps (discovering docs/configs), high semantic search potential (finding CONTRIBUTING.md, CI files) |
170 | 170 | - instruction.md is 2752 chars (well above 500 char preflight threshold) |
171 | 171 | --- |
| 172 | +## 2026-02-16 - US-008 |
| 173 | +- Created onboard-workflow-002: Kafka contributor workflow discovery task |
| 174 | +- Files created: |
| 175 | + - benchmarks/ccb_onboarding/onboard-workflow-002/task.toml (category=workflow_discovery, language=java, difficulty=hard) |
| 176 | + - benchmarks/ccb_onboarding/onboard-workflow-002/instruction.md (6 questions: build prerequisites, Gradle build system, running tests, CI pipeline, code review process, developer workflow example) |
| 177 | + - benchmarks/ccb_onboarding/onboard-workflow-002/environment/Dockerfile (clone apache/kafka@3.9.0) |
| 178 | + - benchmarks/ccb_onboarding/onboard-workflow-002/tests/test.sh (copied from workflow-001, reusable pattern) |
| 179 | + - benchmarks/ccb_onboarding/onboard-workflow-002/tests/ground_truth.json (8 findings, 8 file refs, 2 causal chains, 2 negative checks) |
| 180 | +- Files modified: |
| 181 | + - configs/selected_benchmark_tasks.json (task entry already present, updated total_selected: 198) |
| 182 | + - configs/onboarding_2config.sh (added task ID and SG repo mapping) |
| 183 | +- **Learnings for future iterations:** |
| 184 | + - Kafka uses Gradle build system (NOT Maven or Bazel) — negative checks catch this common misconception |
| 185 | + - Key workflow commands: ./gradlew build, ./gradlew test, ./gradlew :module:test |
| 186 | + - CI uses Jenkins (primary) + GitHub Actions — documented in Jenkinsfile and .github/workflows/ |
| 187 | + - Code review process: Apache JIRA ticket → fork → branch → code → PR → committer review |
| 188 | + - Ground truth for Kafka workflow: 8 required findings (Gradle, build commands, test commands, JUnit, Jenkins, JIRA, PR process), 8 file references (CONTRIBUTING.md, build.gradle, Jenkinsfile, etc.), 2 causal chains (build workflow + PR workflow) |
| 189 | + - instruction.md is 2342 chars (well above 500 char preflight threshold) |
| 190 | + - test.sh verifier continues to be fully reusable across all onboarding task types (orientation, handoff, workflow) |
| 191 | + - MCP benefit score 0.72 for Kafka workflow: high context complexity (multi-module Gradle), moderate cross-file deps (CONTRIBUTING.md, build configs), high semantic search potential (finding docs/configs) |
| 192 | + - The selected_benchmark_tasks.json entry was pre-populated (from infra Ralph) — just needed total_selected count update |
| 193 | + - Negative checks are critical for workflow tasks: prevents claiming wrong build system (Maven/Bazel instead of Gradle) |
| 194 | +--- |
0 commit comments